POST /match

Determine whether pairs of menu item names refer to the same dish, with high accuracy.

Handles misspellings, transliterations, noise, and language differences. "Kadhai Chicken" and "Karahi Chiken" are the same dish. "Butter Chicken" and "Dal Makhani" are not.

Request

Parameter	Type	Required	Description
pairs	object[]	Yes	Array of pairs to compare. 1-100 pairs. Each has text_a and text_b (max 500 chars each).
cosine_threshold	float	No	Similarity threshold for match decision. Default: 0.85.

{
  "pairs": [
    {"text_a": "Kadhai Chicken", "text_b": "Karahi Chiken"},
    {"text_a": "Butter Chicken", "text_b": "Dal Makhani"}
  ],
  "cosine_threshold": 0.85
}

Response

Field	Type	Description
results	object[]	One result per input pair, in submission order

Each result:

Field	Type	Description
text_a	string	First item (as submitted)
text_b	string	Second item (as submitted)
cosine	float	Cosine similarity between embeddings (0-1)
reranker_score	float or null	Precision score (0-1), null if the pair did not pass the cosine gate
match	bool	True if the items refer to the same dish
dietary_conflict	bool	True if one item is veg and the other non-veg
protein_conflict	bool	True if items have conflicting proteins (e.g. chicken vs paneer)

{
  "results": [
    {
      "text_a": "Kadhai Chicken",
      "text_b": "Karahi Chiken",
      "cosine": 0.94,
      "reranker_score": 0.97,
      "match": true,
      "dietary_conflict": false,
      "protein_conflict": false
    },
    {
      "text_a": "Butter Chicken",
      "text_b": "Dal Makhani",
      "cosine": 0.72,
      "reranker_score": 0.12,
      "match": false,
      "dietary_conflict": false,
      "protein_conflict": false
    }
  ]
}

Example

import requests

response = requests.post("https://dish-embed.latimal.com/match",
    headers={"X-API-Key": "YOUR_KEY", "Content-Type": "application/json"},
    json={
        "pairs": [
            {"text_a": "Kadhai Chicken", "text_b": "Karahi Chiken"},
            {"text_a": "Butter Chicken", "text_b": "Dal Makhani"},
            {"text_a": "Cafe Latte", "text_b": "Caffe Latte (Regular)"}
        ]
    }
)

for result in response.json()["results"]:
    status = "MATCH" if result["match"] else "NO MATCH"
    print(f"{result['text_a']} vs {result['text_b']}: {status} (cosine={result['cosine']:.2f})")

Cost

1.0 credits per pair.

Notes

• reranker_score is only populated for pairs that clear the cosine gate. If cosine similarity is very low, the pair is rejected and reranker_score will be null.
• dietary_conflict catches cases like "Classic Burger" vs "Veggie Burger" that look similar but shouldn't be merged.
• protein_conflict catches cases like "Chicken Momos" vs "Paneer Momos".
• For bulk deduplication of an entire menu, use /dedup instead of calling /match for every possible pair.